Block's Dessa voice model rivals human speech
block releases dessa, a full-duplex voice model that sounds human, trained on speech and running locally on gpu. this could change voice ai interactions.
block, led by jack dorsey, has developed dessa, a full-duplex speech model that can converse like a human, with expressions like laughs, pauses, and cries. trained entirely on human speech and running on a single nvidia gpu, dessa aims to provide a more natural interaction. block claims that current voice models sound unnatural due to text-driven speech generation, while dessa builds emotion into the model. the company is seeking safety researchers to test dessa, which can run locally on consumer-grade computers like a macbook m4 max.